Split-Brain Experiments
-
work done by Roger Sperry, Michael Gazzaniga, Joseph
Bogen, P.J. Vogel, Joseph LeDoux, beginning in the 1950s
-
corpus callosum surgically
severed in an attempt to control severe epilepsy
-
left and right hemispheres no longer able to communicate
-
only left hemisphere able to verbally describe images presented
to it
-
both hemispheres able to point to or pick up objects similar
to images presented to it
-
left-handed patient spoke out of left hemisphere after
surgery, but could only write out of right hemisphere
-
another patient developed ability to speak out of right hemisphere
13 years after surgery
Examples
Synthesis
Task: flash different words to different hemispheres and
see if patient can draw an integrated picture
One hemisphere sees SKY, other hemisphere sees SCRAPER,
patient draws picture of sun and clouds next to a scraping knife
Right hemisphere sees FIRE, followed by ARM, patient draws
picture of a rifle
False Memories
Left hemisphere sees picture of a chicken's foot; right
hand (controlled by LH) points to picture of a rooster
Right hemisphere sees picture of a winter snow scene;
left hand (controlled by RH) points to picture of a shovel
Left hemisphere asked why left hand is pointing to a shovel
It doesn't know, but it makes up an explanation anyway:
"left hand chose shovel to clean out a chicken shed"
Narrative Interpreter Mechanism
-
always hard at work seeking meaning of events
-
often makes mistakes by overgeneralizing, constructing potential
pasts as opposed to a true one
-
focus of false-memory research (Phelps, Metcalfe, Funnell)
-
left hemisphere generates many false reports about experiences
-
right hemisphere, in contrast, provides a much more accurate
account (but can't report it through language)
Conclusions
-
enormous degree of lateralization between the hemispheres
-
left hemisphere dominant for language, speech, problem-solving
-
right hemisphere dominant for visual-motor tasks
-
speech and writing are controlled by distinct brain systems
-
left hemisphere serves as an "interpreter" of observed experiences
-
brain/mind is a collection of many quasi-independent modules,
rather than a single problem-solving device
Basic Neurobiology of the Neuron
-
neuron membrane contains ionic channels: embedded
protein structures that act like little gates
-
channels expand or contract in response to chemical and electrical
changes
-
permeability of membrane to ions is a function of the membrane's
electrical polarization and the local chemical concentrations of ions
-
interior of neuron is electrically negative (around -60mV)
with respect to exterior
-
concentration of sodium ions (Na+) much higher outside
neuron
-
concentration of potassium ions (K+) much higher inside
neuron
-
sodium pump maintains concentration gradients; requires
lots of metabolic energy
Action Potentials
-
a negative stimulus further polarizes the neuron membrane
(i.e., makes it more negative); nothing much happens
-
a positive stimulus depolarizes the membrane (i.e.,
makes it less negative)
-
if the amount of depolarization is above some threshold,
the neuron briefly generates a large positive spike
-
spike propagates down the axon to other neurons
-
Hodgkin and Huxley studied axon of giant squid in 1950s;
developed differential equation that predicted quantitative behavior of
axon when stimulated
-
Hodgkin-Huxley equation accurate to within 10% of observed
values
-
has been applied to many other types of neurons
Firing of Action Potentials
-
neuron resting state: K+ roughly in equilibrium; Na+ strongly
out of equilibrium
-
decreasing membrane potential by a small amount may trigger
onrush of Na+ into the cell
-
positive feedback effect: membrane becomes completely permeable
to Na+
-
membrane potential becomes positive (+10s of mV)
-
K+ now strongly out of equilibrium (no longer a strong electrical
gradient; now a strong concentration gradient back out of the cell)
-
K+ ions rush out of the cell
-
electrical imbalances depolarize neighboring regions of membrane
further down the axon
-
spike self-propagates down the axon
-
sodium pump gradually restores ion concentrations; returns
cell to its original state
-
refractory period after firing (threshold elevated; gradually
decreases)
-
in the case of a continuous stimulus, the decreasing threshold
eventually meets the stimulus, causing the cell to fire again
-
a stronger stimulus meets the threshold sooner
-
neuron acts like a voltage-to-frequency converter
-
typical neuronal firing frequencies: 50-100 Hz (up to 500
Hz); background rate 1-5 Hz
Electrical Properties of Axons
-
voltages applied to axons fall off rapidly with distance:
V(x) = V0 e -x/L
-
due to high capacitance of membrane, poor conductance, etc.
-
axons respond slowly to voltage changes: V(t) = 1
- e -t/T
-
axons can't be used to transmit electrical signals directly
-
need to convert to digital signal
-
self-propagating action potential can travel arbitrarily
long distances
-
myelin can speed up action potentials by a factor
of 30
-
"speed of thought": typically up to about 200 ft/sec for
myelinated axons (unmyelinated: ~ 5 ft/sec)
Interactions Between Neurons
-
information transmitted between neurons via chemical process
at synapse
-
at least 50 different neurotransmitters known
-
different types of neurons use different neurotransmitters
-
some important neurotransmitters:
-
GABA (gamma-aminobutyric acid) - inhibitory; used
by 25-40% of all synapses in brain
-
glycine - inhibitory; spinal cord, brain stem (30-40%
of synapses)
-
glutamic acid - excitatory; probably principal excitatory
neurotransmitter in brain; transmitter of granule cells, the most numerous
neurons in the cerebellum
-
histamine - most highly concentrated in brain regions
that regulate emotional behavior
-
substance P - major transmitter of sensory neurons
that convey pain from periphery to spinal cord
-
serotonin - transmitter used in a particular region
of the brain stem
-
dopamine - major transmitter in the brain regions
regulating motor behavior
-
norepinephrine - transmitter of cells in autonomic
(involuntary) nervous system
-
acetylcholine - transmitter for synapses in voluntary
nerve-muscle connections and in many involuntary nervous system synapses
-
many synapses occur on dendritic spines, which are
thought to be the locus of long-term changes in synaptic strength responsible
for many types of learning
-
changes in the shape and geometrical configuration of a spine
can regulate the degree of electrical influence of the presynaptic cell
on the postsynaptic cell
-
changes in the strength of a synapse can occur independently
of other synapses
-
inhibitory and excitatory synaptic influences often combine
in a linear fashion, but plenty of cases in which they do not
-
temporal order of synaptic inputs can have a major influence
on membrane potential (see Chapter 2 for an example)
-
physical location of synapses on dendrites can also have
a major influence
MORAL: Neurons perform very complex analog computations;
their exact behavior depends on a wide range of electrical, chemical, temporal,
and spatial parameters.
Models of the Neuron
McCulloch-Pitts Neuron
-
two possible states: 0 or 1
-
excitatory and inhibitory inputs
-
each neuron has a distinct numerical firing threshold
-
each neuron has a built-in, fixed time delay
-
output is 1 if sum of inputs >= threshold and no inhibitory
input
-
0 otherwise
-
any logical expression can be computed with a network of
McCulloch-Pitts neurons (i.e., M-P neurons are Turing-complete)
-
simpler versions omit time delay, excitatory/inhibitory distinction
Integrate-and-Fire Neuron
-
differential equation relating change in activation to input
signal
-
see Chapter 2 for details
Connectionist Neuron
-
discrete or continuous states
-
output often interpreted as a continuous firing rate (ignores
other information, such as pulse phase)
-
synchronous or asynchronous updates
-
threshold replaced by a more general bias
-
includes activation function (possibly non-linear)
-
some popular activation functions:
-
linear: f(x) = x
-
step: f(x) = +1 if x >= threshold;
0 (or -1) otherwise
-
limiting: f(x) = x if min
<= x <= max; min if x < min;
max
if x > max
-
sigmoid: f(x) = 1 / (1 + e -x)
or tanh(x)
Example
Sigma-Pi Neuron
-
output is weighted sum of conjunctive inputs
-
activations of certain units control the weights
between other units (act as gates)
-
allows for dynamically programmable networks
-
not much work has been done with sigma-pi units to date
Network Architectures
-
feedforward networks: units organized into layers with no
connections within a layer or to earlier layers
-
recurrent networks: any unit may connect to any other unit
(including itself)
-
competitive networks: excitatory and inhibitory connections
within a single layer